A former Bitcoin mining operator disclosed in a June 15 Securities and Exchange Commission filing that it has entered advanced negotiations for a 20-megawatt AI compute contract worth more than $1 billion over its 20-year term — a deal that, if signed, would shut down all Bitcoin mining at its Michigan campus and redirect that power to higher-margin artificial intelligence . The timing is not incidental: North American colocation vacancy has fallen to a record-low 1%, with 92% of capacity currently under construction already committed before a single cabinet is installed, according to JLL's 2026 Global Data Center Market Outlook.

Hyperscale Data, Inc. (NYSE American: GPUS), headquartered in Las Vegas, Nevada, said its subsidiary Alliance Cloud Services expects to execute the Master Services Agreement in the coming weeks. Under the terms under negotiation, the first 10 megawatts would go live within 90 days of signing, with a second 10-megawatt block following 90 days after that. A customer option to expand to 52 megawatts by 2028 would push the cumulative contract value above $2.5 billion — and even then, that full footprint would represent only 17% of the Michigan campus's projected 300-megawatt total capacity.

What Is Driving Enterprises to Independent AI Data Centers?

The search for AI compute capacity that is neither oversubscribed nor controlled by the three major cloud hyperscalers has become one of the defining logistics problems in enterprise technology. Amazon Web Services, Microsoft Azure, and Google Cloud plan to invest more than $500 billion in data center infrastructure in fiscal year 2026, yet demand is outpacing even that construction pace. Data center planning consultant Kirk Killian, speaking at Data Center World 2026, noted that enterprises once planned capacity 12 to 18 months in advance; the current standard is 24 to 36 months.

That scarcity is the structural condition that makes a 300-megawatt campus in Michigan strategically interesting. Power-ready, physically operational sites with existing grid connections are exactly what the industry most urgently needs — and exactly what years of Bitcoin mining investment created.

Why Repurposing a Bitcoin Mine for AI Is Harder Than It Looks

The phrase "repurposing" understates what converting a Bitcoin mining facility into an AI compute campus actually requires. Bitcoin mining runs on application-specific integrated circuits — ASICs — purpose-built chips that perform only one task: solving the SHA-256 hash calculations that validate Bitcoin transactions. Those chips have no role in AI workloads and must be removed entirely and replaced with GPU hardware. A conversion to NVIDIA H100, H200, or AMD MI300X processors also requires installing high-speed interconnects — either InfiniBand fabric or 100-gigabit Ethernet — that standard mining facilities do not carry. Training workloads distribute computation across many GPU nodes simultaneously, and those nodes must communicate at speeds that satellite-grade connectivity cannot support.

The cooling architecture must also change. A Bitcoin ASIC rack draws roughly 1 to 3 kilowatts. A modern GPU rack for AI runs 30 to 132 kilowatts depending on configuration, with NVIDIA's Blackwell GB200NVL72 design requiring approximately 132 kilowatts per rack. Air cooling hits its limits at roughly 70 kilowatts per rack; above that threshold, facilities require direct-to-chip liquid cooling or immersion systems. Industry estimates for a full AI-ready conversion run between $8 million and $11 million per megawatt — meaning a 20-megawatt initial deployment implies a capital commitment in the range of $160 million to $220 million on the customer's side for tech fit-out alone, before the colocation fees.

There is also a critical operational difference between the two workloads. Bitcoin mining can be interrupted without consequence — operators can shut down thousands of ASICs in seconds when power prices spike, which is why miners are valuable to grid operators as flexible load. AI training workloads cannot be paused mid-run without losing the computation already performed. That means AI data centers need uninterruptible power with redundant backup systems, rather than the flexible load profile that mining facilities historically offered utilities.

Hyperscale Data said in May 2026 that it had already ordered and paid for key electrical and infrastructure equipment to support rapid deployment. The company has also been fitting out sections of its existing building and agreed in March 2026 to acquire an additional 48.5 acres adjacent to the campus, more than doubling its acreage in the area.

The Deal Structure and What It Does Not Yet Guarantee

The company's June 15 SEC filing is careful to note that no definitive agreement has been signed. Under the terms under negotiation, Hyperscale Data's subsidiary Alliance Cloud Services would deliver colocation and related data center services. Chief Executive Officer Will Horne expressed confidence in the company's trajectory but did not name the prospective customer. The scale of the negotiations — a multi-decade, multi-megawatt agreement — points to a hyperscale-tier AI operator or a large enterprise with substantial training or inference requirements, though neither the customer's identity nor the final contract terms have been confirmed.

The $2.5 billion figure that has attracted investor attention applies only if the unnamed customer elects to expand its footprint to 52 megawatts by 2028, subject to the company delivering the additional 32 megawatts of capacity it has outlined. The company cautioned that the campus expansion beyond the initial 20-megawatt discussion remains subject to regulatory approvals, utility agreements, engineering studies, and financing — none of which are in place.

A Federal Lawsuit Is Already Pending Against the Same Subsidiary

The subsidiary named in the AI compute negotiations — Alliance Cloud Services — is also the defendant in a federal class-action lawsuit filed in the U.S. District Court for the Western District of Michigan in late May 2026. Two residents living within one mile of the company's existing Dowagiac, Michigan, data center filed suit alleging that the facility generates constant, excessive industrial noise from its cooling systems, operating 24 hours a day, seven days a week. The complaint states that Alliance Cloud Services has failed to implement adequate soundproofing equipment and that the noise has disrupted the use of nearby homes and affected property values.

Laura Sheets, lead attorney at Liddle Sheets, P.C., said in a statement: "These people had no say in the introduction of this large and loud data center into their otherwise peaceful community." Alliance Cloud Services did not respond to a request for comment cited in news reports of the lawsuit.

Dowagiac Mayor Patrick Bakeman had already signaled friction with the company's expansion plans before the lawsuit was filed. In an April 1, 2026, open letter to CEO Will Horne, the mayor stated that Alliance Cloud Services had not yet applied for or received the regulatory approvals needed to move forward with its planned expansion. The city subsequently revised its noise ordinance in March 2026 to set explicit limits on sound levels during the day and night.

The Dowagiac facility and the Michigan campus central to the new AI compute negotiations are both operated under Alliance Cloud Services, though the sites serve different operational functions. Both fall under the subsidiary that prospective AI customers would be contracting with — a legal exposure that any enterprise evaluating a 20-year commitment would be expected to assess.

What Investors Are Watching

GPUS shares rose more than 24% in pre-market trading following the June 15 announcement. That reaction reflects the significance of a credible $1 billion contract in advanced negotiations for a company operating at a much smaller scale. Context is warranted: the stock was trading at approximately $0.15 per share heading into the announcement, down roughly 96% over the prior year, and its short-term liabilities exceeded liquid assets as of the most recent analysis available.

Hyperscale Data has issued a series of positive AI-related announcements over the past several months — including revenue guidance of $180 million to $200 million for 2026, disclosed in March, and the announcement of humanoid robot production beginning at the Michigan campus on June 11. Prior AI-tagged news releases for GPUS averaged a next-day stock movement of approximately -0.33%, according to market history analysis, suggesting that investor response to announcements without signed agreements has been cautious.

The company has structured its corporate future around the Michigan campus. It expects to divest its Ault Capital Group subsidiary in the second quarter of 2027, after which it intends to operate primarily as a data center business focused on high-performance computing. That planned transformation depends materially on whether the current negotiations produce a signed agreement.

Frequently Asked Questions

What is AI data center colocation?

AI data center colocation is a model in which a company rents physical space, power, and cooling infrastructure from a third-party facility operator and installs its own GPU hardware to run AI workloads. Unlike public cloud services, colocation gives enterprises direct control over their hardware while outsourcing the facility, power delivery, and cooling systems. AI colocation differs from traditional colocation in its power density requirements: standard server racks draw 3 to 5 kilowatts, while GPU-equipped AI racks routinely require 30 to 132 kilowatts per rack.

Why are enterprises seeking AI compute capacity outside AWS, Azure, and Google Cloud?

North American data center colocation vacancy has fallen to a record-low 1%, with 92% of capacity currently under construction already pre-committed before a single cabinet is installed, according to JLL's 2026 market data. The top three hyperscalers are investing more than $500 billion in data center infrastructure in fiscal year 2026, yet demand continues to outpace construction timelines. Enterprises that once planned capacity 12 to 18 months out now need 24 to 36 months of lead time. Former Bitcoin mining facilities — which already have large grid connections, operational cooling infrastructure, and physical space — offer a faster path to energized AI capacity than new construction.

Can a Bitcoin mining facility actually run AI workloads without major changes?

No. Bitcoin mining uses ASIC chips designed exclusively for SHA-256 hashing, which must be fully replaced with GPU hardware. The networking infrastructure must also be overhauled to support high-speed interconnects — InfiniBand or 100-gigabit Ethernet — required for distributed AI training. Cooling systems must be upgraded to handle 10 to 40 times the power density per rack. Industry estimates for a full conversion run between $8 million and $11 million per megawatt of capacity.

What risks does Hyperscale Data face before the Michigan deal closes?

No agreement has been signed as of the company's June 15, 2026, SEC filing. The full expansion to 300 megawatts remains subject to regulatory approvals, utility agreements, financing, and engineering studies. A federal class-action lawsuit is pending against Alliance Cloud Services — the same subsidiary named in the AI compute negotiations — alleging chronic noise pollution at its existing Dowagiac, Michigan, facility. The company's stock traded at approximately $0.15 per share ahead of the announcement, down roughly 96% over the prior year, and its short-term liabilities exceeded liquid assets as of the most recent available data.